When Averages Lie: Mean, Median, and the Story They Hide

From Tukuk-OS Wiki, the free knowledge base.

"Average house price," "average salary," "average rainfall" — the same word can describe three very different numbers, and picking the wrong one changes every conclusion. A plain-language course on mean, median, distribution and the art of reading a statistic without being read by it.

Three Ways to Be "Typical"

Every number with the word "average" in front of it is secretly a choice between three different questions. The mean answers: if the total were shared equally, what would each get? Add everything, divide by the count — the definition of arithmetic fairness, and the most useful single number when the data is well-behaved. The median answers: what does the person in the exact middle experience? Sort the list, take the centre value; half the population is above, half below, and the extremes have no power over it whatsoever. The mode answers: what is the most common single value — the only "average" that works for categories, and the one that matters when the crowd wants to know what typical actually looks like.

In a symmetric, well-behaved distribution these three coincide, which is why everyday intuition treats them as one number. The trouble is that real life is full of distributions where they part company — and every consequential statistic ever published lives in that gap.

The House Price Problem

Take a street of ten houses: nine valued around 300,000 and one at 3.2 million. The mean is 590,000 — a number that describes exactly one resident and misrepresents the other nine. The median is 300,000, which is what "typical" means to almost anyone standing on that pavement. A developer announcing "average values rising along the corridor" is quoting the mean; a resident asking what the neighbourhood is like gets the honest answer from the median — and both are technically correct, which is the whole lesson. Income statistics are the same trap with bigger teeth: because a small minority holds a long tail of enormous values, mean income sits persistently above median income in every country, and the gap between the two is itself a measure of inequality. When salaries are quoted as means in job adverts, the number is the version of the average that flatters the employer's side; the median is the number to ask for, and its absence tells you something.

Rainfall uses the same maths against farmers: a region whose "average annual rainfall" is normal can be built from two cyclone years and eight droughts — a mean that hides bimodality, the distribution where the mode is honest and the mean is a fiction no year ever matched.

The Rules That Keep You Honest

First rule: know the shape before trusting the mean. A distribution's skew (the lean toward one long tail) and its spread (standard deviation, or the more robust interquartile range) decide how representative the mean can be. Income is right-skewed, test scores are often left-truncated, human heights cluster — and a mean quoted from skewed data without the median beside it deserves an asterisk. Second rule: watch the denominator. "Average deaths per year" can be padded by changing the base period; "1 in 4 statistics students" can shrink the population to make a rate look alarming. Third rule: ask who was excluded. Survey averages quietly summarise only the people who answered; non-response bias has an official name because the effect on published means is official too, and "average" in a headline often really means "average of respondents." Fourth rule: small samples bounce. A two-person average tells you nothing about a population, and the smaller the sample behind a quoted mean, the less the third decimal place means — a discipline everyday percentage maths shares with shopping and statistics alike.

Where does the mean remain the right tool? Two places especially: when the total is what matters — energy consumption, crop yield, reservoir inflow — because mean times count equals total exactly, and no other average gives you that identity. And when the distribution is genuinely symmetric, where the mean is the most statistically efficient estimate, using every data point rather than only the middle rank. The failure mode is neither of these: the skewed, the fat-tailed, the bimodal — which is to say most of what governments, markets, and journalists report.

The habit to keep is worth more than the definitions: whenever an "average" appears in the wild, ask which one it is, what the middle person actually gets, and what the shape looks like before the number did. Most statistics are not lies — they are honest numbers answering a different question than the one you wanted asked. The mental-math toolkit helps you do the reconstruction in your head, in real time, without trusting the person holding the spreadsheet.

See also

• Weather vs Climate — The Real Difference, and Why the Confusion Costs Us
• Percentages, Ratios and Discounts: The Mental Maths of Everyday Life
• How Countries Count Their People — Census, Sampling and the Numbers Behind the Numbers